Ultracortex + Thinkpulse, LSL, and EEGlab / import and noise issues
Hello,
Currently, my lab is struggling to import data from our Ultracortex 8 electrode headset with think pulse electrodes. We are collecting data via networking openbci.gui with the labrecorder which exports data into .xdf file type. When importing data into eeglab via the .xdf import extension MatLab displays a warning to the effect of, "channel labels missing, generating new channel labels". Then, when viewing our data eeglab displays an output containing 125 channels and data that does not appear similar to what we see in openbci.gui.
Any help or suggestions would be greatly appreciated.
Comments
Hi Tbuck,
You may want to make an inquiry on the EEGLAB discussion forum. This may be an interoperability issue between LabStreamingLayer recorder and EEGLAB.
https://labstreaminglayer.readthedocs.io/info/eeglab.html
https://eeglab.org/others/EEGLAB_mailing_lists.html
Mentioning Richard @retiutut.
Regards, William
Hi William/ @wjcroft,
I appreciate your response and I have and will continue to reach out via those venues as well as here. We are still having an issue importing our data. Arnaud Delorme from the EEGlab boards brought up a concern about the drivers used with the EEG headsets being at issue, I am curious what your opinion on this is.
This aside we were able to achieve slightly better results via the python code for obtaining an lsl stream of our ultracortex, however while the data imports to eeglab as a labeled 16 electrode EEG the data itself is relatively illegible (in comparison to other EEGlab data we've seen)/ not quite what we see in openbcigui (please see the attached screenshot for reference). Any thoughts or comments you or others have would be greatly appreciated.
Sincerely,
Tim
Just offhand I observe that the graph above has TONS of mains noise, the very high frequency waveforms. Have you applied a notch filter at your mains frequency? Also suggest a bandpass filter at say .5 Hz to 45 Hz.
Position your EEG equipment away from mains noise sources, such as extension cords, power supplies, LED lamps, conduits (in walls, floors, ceilings), Wifi and cellular phone equipment.
I applied the notch filtering and bandpass filters in subsequent trials after receiving your most recent message. I also attempted to use my equipment in different environments ( e.g. my lab room vs a study room), received the following data
In lab room
In study room with lights on
In study room with lights off
It does appear that the mains noise is a factor in our illegible data, however, would that also account for the amplitude of the waveforms throughout and the seeming traveling that they do (especially evident in image 1). If this is the case would altering the gain of the electrodes as well as seeking a "quieter" environment be our potential end solution?
Tim,
It's not clear to me that your mains notch filtering is working. Because both image 1 and 2 still show the strong high frequency mains waveforms. Although the amplitude of the mains noise HAS decreased compared to your original Oct 6 post image. Most filter function parameters have both a frequency (or range), AND a "filter order", which is how sharp or deep the filter is. You may try increasing the filter order on the notch.
But 3rd image does imply that a primary source of your mains noise is from the lighting. Since standing wave mains junk is strongly reduced in the 3rd image. HOWEVER, the mains can still be seen in image 3, if you look closely. So I would suspect other sources than the lighting.
The "moving baseline" aspect of your images is usually because of very low frequency components, caused by the scalp DC offset potential slowly moving over time. That was the intent of using the bandpass to cutoff components below .5 Hz. Either the filter order of your bandpass is relatively weak, or you might benefit from moving the .5 Hz up to 1 Hz, to increase the DC offset cutoff.
Image 1 has particularly strong DC drift, and I am wondering if this was recorded in the first few seconds of starting the sampling. Any electrodes take a few seconds to stabilize; so you always need to discard the first seconds of the recording.
Note on the 3rd image in the last post, there is STILL one channel showing HUGE mains noise, where all other channels have been attenuated. Not sure what is going on there, but it could signify poor scalp contact on that channel.
I am mentioning @julienConscious, the ThinkPulse developer, in case he might have any other comments or insights.
Best regards, William
Hi Tim, William,
signal looks very noisy indeed, slightly better in the last image (except for one or 2 noticeable channels). Looks like the Y scale is roughly 10 microV so this is a decent amplitude. What's unusual is that there is a pulsating noise that does not look like mains noise that typically creates steady high amplitudes/high frequencies.
I believe you mentioned that it is very different from what you see in the OpenBCI GUI. Could you send an image from the GUI where we see the times series and the Spectrum. Do you have 50 or 60Hz mains noise?
Before exploring the possible contact problems, we should make sure the LSL and the filters are not the problem and I think using the OpenBCI GUI is the best way to make sure the original signal is similar to the signal you see in Python or in EEGLab.
I suggest you try a 5-50Hz band pass. Also, in the GUI try deactivating the "bad" channels (if you use Bias, they contaminate the other ones). You should obtain a very stable signal centered around zero, blinks should be clearly visible on Fp1/2 and jaw clenching should create a high amplitude artefact on pretty much all channels. Eventually, when closing the subject's eyes, you should see alpha wave particularly on O1/2.
Best,
Julien
Julien, thanks.
Tim, if you make some screenshots with the GUI, be sure to experiment with turning off / on the various filters and notch, so you can get an idea of the look before and after filters.
While the 'pulsing' beating noise is not typical mains noise (that is more constant as Julien says), it still looks suspiciously like the high frequency component is near mains 50 or 60 Hz. So there may be multiple frequencies 'beating' together to produce these standing wave pulses. The fact that the EEG became much less noisy with the lights off, is also a big clue that this is mains related.
In the last image you posted, you can STILL see these high frequency noise waveforms if you look closely. So it is not only coming in with the lights.
William